The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill researches startup URLs, repositories, and landing pages to extract business data. This creates a surface for indirect prompt injection where a malicious website could attempt to influence the agent's analysis or output.
Ingestion points: Startup URLs, repository content, and competitor pricing pages are parsed for data as part of the core research workflow (SKILL.md, Workflow Step 1 & 2).
Boundary markers: The skill instructs the agent to "Distinguish verified product facts, public market evidence, user-supplied numbers, working assumptions, and unknowns," which serves as a natural boundary for evaluating the quality of ingested data.
Capability inventory: The skill executes bundled Python and Node.js scripts to process data and writes findings into the user's workspace (SKILL.md, Steps 5, 6, 7).
Sanitization: The skill implements strict data handling: scripts/generate_business_plan.py includes a safe_url function to prevent markdown link injection, and scripts/prepare_plan.py validates all extracted data against a rigid JSON schema before it is used for financial calculations or report generation.
[COMMAND_EXECUTION]: The skill uses local subprocess calls to execute its own bundled scripts for data validation and report generation.
Evidence: SKILL.md contains instructions to run python3 "$SKILL_DIR/scripts/prepare_plan.py", python3 "$SKILL_DIR/scripts/generate_business_plan.py", and node "$SKILL_DIR/scripts/generate_financial_model.mjs".
These executions are limited to the skill's internal logic and are used to provide the primary functionality of generating business plans and spreadsheets.

startup-business-planner